The learning system includes a data generation unit configured to generate learning data based on rehabilitation data and a learning unit configured to perform machine learning using the learning data. A sensor is provided to detect a plurality of motion amounts in a walking motion of a trainee, and it is evaluated that, when one of the motion amounts matches one of abnormal walking criteria, that the walking motion is an abnormal walking pattern that meets the matched abnormal walking criterion. The data generation unit generates each of the pieces of rehabilitation data before and after a change in the results of evaluation of the abnormal walking pattern as learning data. The learning unit sequentially inputs each of the pieces of rehabilitation data as one data set, thereby performing machine learning.
Legal claims defining the scope of protection. Each claim is shown in both the original legal language and a plain English translation.
2. The learning system according to claim 1, wherein it is determined whether the walking motion of the trainee matches the abnormal walking criteria using an average value of motion amounts for a plurality of walking cycles after the change in the setting value of the setting parameter.
This invention relates to a learning system for analyzing and correcting abnormal walking patterns. The system monitors a trainee's walking motion and compares it against predefined abnormal walking criteria. The system adjusts a setting parameter to modify the trainee's walking motion and evaluates the effect of this adjustment. To determine whether the walking motion matches the abnormal criteria, the system calculates an average value of motion amounts across multiple walking cycles after the parameter adjustment. This allows the system to assess whether the correction has been effective by analyzing consistent motion data over time. The system may include sensors or other devices to measure walking motion, such as joint angles, step length, or gait symmetry. The abnormal walking criteria may define thresholds or patterns indicative of gait disorders or inefficiencies. By averaging motion data over multiple cycles, the system reduces noise and variability, providing a more reliable assessment of the trainee's walking improvement. The invention aims to improve rehabilitation or training programs by objectively evaluating walking corrections.
3. The learning system according to claim 1, wherein, when setting values of two or more of the aforementioned setting parameters change at the same time in consecutive walking cycles, the learning unit weights each of these two or more of the aforementioned setting parameters changed at the same walking cycle.
A learning system for robotic walking mechanisms addresses the challenge of optimizing multiple gait parameters simultaneously to improve stability and efficiency. The system includes a learning unit that adjusts setting parameters controlling the robot's walking behavior. When two or more of these parameters are modified in the same walking cycle, the learning unit assigns weights to each parameter to determine their individual contributions to the observed walking performance. This weighting mechanism helps distinguish the influence of each parameter when multiple changes occur concurrently, enabling more precise adjustments in subsequent cycles. The system may also include a parameter setting unit that initializes or updates the parameters based on predefined rules or external inputs, and a walking unit that executes the walking motion according to the current parameter values. By analyzing the effects of simultaneous parameter changes, the system refines the walking behavior iteratively, improving adaptability to different environments and tasks. The approach is particularly useful for legged robots operating in dynamic or unstructured settings where real-time adjustments are necessary.
5. The learning method according to claim 4, wherein it is determined whether the walking motion of the trainee matches the abnormal walking criteria using an average value of motion amounts for a plurality of walking cycles after the change in the setting value of the setting parameter.
This invention relates to a learning method for training individuals to recognize abnormal walking patterns. The method addresses the challenge of accurately detecting deviations in walking motion, which is critical in applications such as medical rehabilitation, biomechanics, and assistive robotics. The system involves adjusting a setting parameter that influences walking motion, then analyzing the trainee's walking cycles to determine if the motion matches predefined abnormal walking criteria. The determination is based on an average value of motion amounts across multiple walking cycles after the parameter adjustment. This approach ensures robustness by averaging over multiple cycles, reducing the impact of transient variations. The method may also include generating a learning signal based on the comparison between the trainee's motion and the criteria, which can be used to guide further training or adjustments. The system may incorporate feedback mechanisms to refine the parameter settings iteratively, improving the accuracy of abnormal walking detection over time. The invention is particularly useful in scenarios where precise identification of abnormal gait patterns is necessary for therapeutic or diagnostic purposes.
6. The learning method according to claim 4, wherein, when setting values of two or more of the aforementioned setting parameters change at the same time in consecutive walking cycles, the machine learning weights each of these two or more of the aforementioned setting parameters changed at the same cycle.
This invention relates to a machine learning method for optimizing walking parameters in a bipedal robotic system. The problem addressed is the challenge of efficiently adjusting multiple walking parameters simultaneously in a robotic walking cycle to improve stability and performance. Traditional methods often adjust parameters sequentially, which can be inefficient and may not account for interactions between parameters. The method involves a machine learning system that monitors and adjusts two or more walking parameters in the same walking cycle. These parameters may include joint angles, gait timing, or other motion-related variables. When multiple parameters are changed at the same time, the machine learning model assigns individual weights to each parameter to evaluate their combined impact on walking performance. This allows the system to assess how different parameter combinations influence stability, energy efficiency, or other walking metrics in real time. The weighted adjustments help the robot adapt more dynamically to varying conditions, such as uneven terrain or changes in payload, by learning optimal parameter interactions. The approach improves the robot's ability to maintain balance and efficiency without requiring extensive trial-and-error testing.
7. A non-transitory computer readable medium storing a program for causing a computer to execute the learning method according to claim 4.
This invention relates to machine learning systems, specifically a method for training a model to improve its performance by adjusting hyperparameters based on validation data. The problem addressed is the inefficiency of traditional hyperparameter tuning, which often relies on manual trial-and-error or computationally expensive grid searches. The solution involves an automated process where a model is trained using a set of initial hyperparameters, then evaluated on validation data. Based on the evaluation results, the hyperparameters are adjusted iteratively to optimize model performance. The method includes steps for selecting an initial set of hyperparameters, training the model, evaluating its performance, and updating the hyperparameters using an optimization algorithm. The process repeats until a stopping criterion is met, such as reaching a maximum number of iterations or achieving a desired performance threshold. The invention also includes a non-transitory computer-readable medium storing a program that executes this learning method, enabling efficient and automated hyperparameter tuning for machine learning models. This approach reduces the time and computational resources required compared to traditional methods while improving model accuracy.
Cooperative Patent Classification codes for this invention. Click any code to explore related patents in that topic.
May 28, 2020
April 2, 2024
Browse 5M+ US patents with plain-English claim translations and AI-generated analysis.